Role of the basal ganglia in conditional associative learning : a multidisciplinary approach

HAL Id: tel-00006160

https://tel.archives-ouvertes.fr/tel-00006160

Submitted on 27 May 2004

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

learning : a multidisciplinary approach

Fadila Hadj-Bouziane

To cite this version:

Fadila Hadj-Bouziane. Role of the basal ganglia in conditional associative learning : a multidisciplinary approach. Neurons and Cognition [q-bio.NC]. Université Claude Bernard - Lyon I, 2003. English. �tel-00006160�

UNIVERSITE CLAUDE BERNARD - LYON 1 Thèse d'Université

Mention Neurosciences

soutenue le 17 Décembre 2003

'Role of the basal ganglia in

conditional associative learning :

a multidisciplinary approach'

'Rôle des ganglions de la base dans l'apprentissage associatif conditionnel : une approche multidisciplinaire'

Fadila HADJ-BOUZIANE

Directeurs de thèse :

Dr. Driss BOUSSAOUD - Dr. Martine MEUNIER

Pr. Marc Jeannerod, Président Pr. Bioulac, rapporteur

Dr. Apicella, rapporteur Pr. Dubois, examinateur Dr. Burnod, examinateur Dr. Driss Boussaoud

___________________________________________________________________________ RESUME en français

Avec l'expérience, nous acquérons une panoplie de règles, associations arbitraires entre des stimuli externes et des actes moteurs, qui nous permettent d'adapter notre comportement à l'environnement (apprentissage associatif conditionnel). Ce type d'apprentissage met en jeu les boucles reliant les ganglions de la base (GGB) et le cortex frontal. Ce travail visait à préciser le rôle des GGB dans l’apprentissage de règles visuo-motrices conditionnelles en utilisant plusieurs approches : 1) l’enregistrement de l’activité des neurones du striatum chez le singe éveillé, 2) l’étude chez des patients atteints de la maladie de Parkinson (une pathologie neurodégénérative touchant les GGB) et 3) la neuroimagerie fonctionnelle chez l'homme sain. Les résultats des trois expériences convergent pour indiquer que les GGB sont impliqués à la fois dans l'acquisition et la rétention des associations visuo-motrices.

Role of basal ganglia in conditional associative learning : a multidisciplinary approach ___________________________________________________________________________ RESUME en anglais

The arbitrary mapping of sensory information onto action forms an important element of the intelligent behavior of primates (also called conditional associative learning). The cortico-basal ganglia-thalamo-cortical loops are thought to play a key role in such behavior. The present research was undertaken to investigate the role of the basal ganglia (BG) in conditional visuo-motor associative learning using three complementary approaches: 1) single-unit recordings in awake monkeys, 2) behavioral testing in patients suffering from Parkinson's disease (a neurodegenerative disease affecting the BG), and 3) functional neuroimaging in healthy subjects. The results of all three studies converge to indicate that the BG are involved in both the acquisition and the retention phases of visuo-motor associations. ___________________________________________________________________________ DISCIPLINE : Neurosciences

___________________________________________________________________________

MOTS-CLES

Ganglions de la base, striatum, système fronto-striatal, apprentissage visuo-motor conditionnel, neurophysiologie, singe, neuropsychologie, Maladie de Parkinson, IRMf, approche multidicsiplinaire

KEY WORDS

Basal ganglia, striatum, fronto-striatal system conditional visuo-motor learning, neurophysiology, monkeys, neuropsychology, Parkinson's Disease, fMRI, multidisciplinary approach

___________________________________________________________________________ INTITULE ET ADRESSE DE L'U.F.R. OU DU LABORATOIRE :

Institut des Sciences Cognitives UMR 5015, CNRS-UCBL

A. INTRODUCTION

SECTION 1 - BASAL GANGLIA ANATOMY

...14

I. THE BASAL GANGLIA CONCEPT : HISTORICAL EVOLUTION

...14

II. THE BASAL GANGLIA COMPONENTS

...17

1. The striatum ... 17

a. Anatomical subdivisions ... 18

b. Cytology... 18

c. Functional domains : matrix/striosome compartments ... 23

2. The Globus pallidus (GP)... 24

3. The substantia nigra (locus niger)... 25

4. The Subthalamic nucleus (Luys Body) ... 25

III. INPUTS TO THE BASAL GANGLIA AND 'THE BASAL GANGLIA LOOPS'

...27

1. The cortico-striatal projections : a funneling or a parallel processing? ... 27

a. Kemp & Powell's proposal ... 27

b. Alexander, Delong and Strick's proposal ... 28

c. Parent's poposal ... 32

2. The Nigrostriatal projections... 32

3. The Thalamostriatal projections ... 34

4. Amygdalostriatal projections ... 34

5. Other sources of striatal inputs ... 35

6. Integration by striatal neurons of different inputs ... 35

IV. OUTPUT OF THE BASAL GANGLIA AND 'THE BASAL GANGLIA LOOPS'

...37

V. BASAL GANGLIA INTRINSIC CIRCUITS: THE DIRECT AND INDIRECT PATHWAYS

...

...38

VI. INFORMATION PROCESSING IN THE BASAL GANGLIA: A RE-EVALUATION OF THE CLASSICAL MODEL

...40

2. Direct/indirect model?... 40

3. Information processing in the basal ganglia: feedfoward/feedback -parallelism/convergence? ... 41

4. New perspectives? ... 42

a. Joel and Weiner model: the “split circuits" ... 42

b. Striatal compartments ... 42

SECTION 2 - CONDITIONAL VISUO-MOTOR LEARNING IN PRIMATES : A KEY ROLE FOR THE BASAL GANGLIA

...46

I. ROLE OF THE FRONTO-STRIATAL SYSTEM IN CONDITIONAL VISUOMOTOR ASSOCIATIONS

...47

1. The frontal cortex: brief anatomical description... 47

2. Role of the frontal cortex in conditional visuomotor associations... 48

a. Neuropsychology in humans and monkeys ... 48

b. Brain imaging in humans and neurophysiology in monkeys... 50

3. The basal ganglia and conditional visuomotor associations ... 51

a. Neuropsychology in humans and monkeys ... 51

b.Brain imaging in humans and neurophysiology in monkeys... 52

II. LINKING SENSORY INFORMATION TO MOTOR RESPONSES: A SPECIFIC ROLE FOR THE STRIATUM

...54

1. The striatum: a site of convergence for sensory, motor, and reward signals ... 54

2. Coding for stimulus versus movement in frontal cortex and striatum... 55

III. A MODEL FOR DISTRIBUTED PROCESSING IN THE FRONTO-STRIATAL SYSTEM DURING LEARNING

...61

1. Lateral prefrontal cortex (PFdl and PFvl) ... 61

2. The dorsal premotor cortex (PMd)... 62

3. The striatum ... 63

B. NEUROPHYSIOLOGICAL STUDY IN MONKEYS

...66

3. Surgery and recordings ... 67

4. Data analysis... 68

III. RESULTS

...70

1. Behavior ... 70

2. Neuronal activity... 72

a. General properties of the striatal neurons... 72

b. Modification of striatal activity during learning... 77

IV. DISCUSSION

...89

1. Summary of the principal findings ... 89

2. General properties of striatal neurons during the execution of well-learned arbitrary visuo-motor associations... 89

3. Modulation of activity in the striatum during learning of novel visuo-motor arbitrary associations: a comparison with changes in the frontal lobe... 90

4. Alternative explanations ... 91

5. Limits ... 92

C. NEUROPSYCHOLOGICAL STUDY IN PARKINSON

PATIENTS

...94

II. SUBJECTS AND METHODS

...98

1. Subjects... 98

a. Patients ... 98

b. Controls... 98

2. Materials ... 102

3. Procedure ... 102

Standard mapping task (SM)... 102

Single association learning without working memory (SLnoWM)... 103

Single association learning with working memory (SL_WM)... 103

Visuo-motor Conditional associative learning task (CAL) ... 103

4. Data analysis... 106

2. Single association learning, without (SLnoWM) or with (SL_WM) working memory.107

3. Visuo-motor Conditional Associative Learning ... 109

a. Comparison between the controls and PD-I OFF medication ... 109

b. Comparison between the control and PD-II OFF medication ... 109

d. L-Dopa treatment effect ... 110

IV. DISCUSSION

...114

1. Summary of the results... 114

a. Preserved performance on standard mapping and single association learning tasks ... 114

b. Conditional associative learning impairment in a majority of PD patients ... 114

2. Comparison with earlier studies ... 116

3. Basal ganglia and CAL... 117

D. NEUROIMAGING STUDY IN HEALTHY HUMANS

...120

II. MATERIALS AND METHODS

...122

1. Subjects and setup... 122

2. Behavioral paradigms ... 123

3. Testing procedure ... 123

4. Behavioral analysis ... 124

5. MR acquisition ... 124

6. Image processing and statistical analysis... 125

III. RESULTS

...128

1. Behavioral data ... 128

2. fMRI activation data: Early learning of novel versus execution of familiar stimulus-response or location-stimulus-response associations... 130

3. fMRI activation data: Early versus late learning of novel stimulus-response or location-response associations... 131

IV. DISCUSSION

...139

3. Subcortical activation ... 142 4. Limits ... 143

E. GENERAL DISCUSSION

1. Electrophysiological study : results and perspectives ... 145 2. Functional imaging study : results and perspectives ... 147 3. Neuropsychological study : results and perspectives ... 147 4. What does conditional associative lerning tell us about basal ganglia functions ?. 148

In the typical course of daily events, we make a variety of body movements on the basis of what we sense in our environment. Often, we gaze at an object present in our peri-personnal space (e.g. a cup of coffee), attend to its features and place, reach toward it, and grasp it. Such movements were termed by Wise and colleagues (1996) standard sensorimotor

mapping in that the movement is mapped accurately onto the target of action. The brain uses

the location of the object to guide the hand through space, and the shape, size and texture of the object to form the appropriate grasp (Jeannerod, 1997). This type of visuomotor transformations relies on direct, cortico-cortical connections linking the occipito-parietal visual pathway (dorsal visual stream), which processes visuospatial information, to the frontal motor and premotor regions, which control the selection, planning and execution of voluntary movements (Figure O1)

However, mammals in general, and primates in particular, perform far more than simple standard movements. Through evolution, the brain has developed a tremendous capacity to link sensory information to motor responses through purely arbitrary rules. In humans, this non-standard mapping (Wise et al., 1996) is present in numerous everyday activities. Abilities such as car driving and phone handling depend on it, as do many language-related skills. We have all learned to stop at a red traffic light and to go at a green one, or to wait for a specific tone before dialing a phone number and to hang up when hearing a busy signal. Likewise, reading is based on learned relationships between the visual form of letters and the movements necessary to pronounce them. Arbitrary sensorimotor associations are also of highly adaptive value for nonhuman primates living in their natural habitat. For example, African vervet monkeys learn through experience to select an escape response according to the specific sound of their conspecifics' alarm calls. Schematically, one sound instructs to stand up, peer into the surrounding grass and watch for a snake, another, to flee into trees away from a leopard, and still another, to run into bushes to hide from an eagle (Cheney & Seyfarth, 1990).

Understanding how arbitrary sensorimotor associations are learned, and how they are retrieved and used when the context requires them, has been one of the challenging issues for cognitive neuroscience. Experimental tasks have therefore been designed in order to assess this type of associations in laboratory situations. Generally, these experimental tasks use two

displacements, etc.). Subjects, whether human or nonhuman primates, are required to learn and then execute arbitrary rules such as 'if green go right, if red go left'. Hence, these tasks are often referred to as 'conditional' associative tasks. A noteworthy particularity of conditional tasks is that all stimuli being equally associated with reward (or success), correct responses cannot be driven by simple stimulus-reward associations (i.e. approach the rewarded item or class of items, and avoid the non-rewarded one). Instead, subjects must link a stimulus with a response which in turn leads to reward. In their vast majority, experimental studies have focused on how visual stimuli are mapped onto motor responses, in part because the brain organization of vision is better known than that of other sensory modalities. A few experiments on auditory-motor associations suggest, however, that results obtained for vision could apply to other modalities as well.

Figure O1. Schematic representation of the cerebral substrates of standard (blue) and

non-standard (red) sensorimotor mapping. The former relies on the dual parieto-premotor pathway controlling reaching and grasping, whereas the latter involves a more complex network centered on the loop linking the premotor cortex to the striatum.

Striatum Prefrontal Premotor Inferior Temporal Parietal Globus Pallidus Thalamus

associations has relied on a combination of four main approaches, human neuropsychology in brain-damaged patients, experimental lesions in monkeys, imaging studies in healthy human subjects, and electrophysiological recordings in intact, awake monkeys. Valuable insights have been gained that indicate that non-standard mapping involves a complex brain network through which the posterior sensory cortices (in particular the dorsal and ventral visual streams), but also the prefrontal cortex, the hippocampal region, and possibly the cerebellum, interact with the loop linking the lateral premotor cortex to the basal ganglia, via the thalamus (Figure O1). Much remains to be done, however, to fully understand the specific contribution of each of this network components to the complex processes underlying arbitrary sensorimotor associations.

Four years ago, at the time the present project was initiated, available data provided strong evidence that the dorsal portion of the lateral premotor cortex (PMd), a region well-known for its role in motor preparation, plays an important role in arbitrary visuomotor associations. Briefly, damage to PMd in humans (Halsband & Freund, 1990) and monkeys (e.g. Halsband & Passingham, 1985) had been found to profoundly disrupt both the acquisition of new associations and the execution of well-learned ones. In addition, single-cell recordings had not only revealed neural properties in PMd cells likely to reflect the selection of action in response to sensory cues in over-trained monkeys (Boussaoud & Wise, 1993a,b), but had also demonstrated the existence, within PMd, of a learning-related plasticity in animals engaged in the acquisition of novel associations (Mitz et al., 1991). By contrast, knowledge regarding the role of the basal ganglia, and in particular, of its main input structure, the striatum, which is intimately linked with PMd, was scarce. A few neuropsychological studies of patients suffering from Parkinson's disease, one of the main pathologies affecting the basal ganglia, had provided contradictory findings as to whether or not these patients remained able to learn conditional associative tasks (e.g. Gotham et al., 1988; Pillon et al., 1998), and, among them, only one had specifically addressed the issue of sensorimotor (as opposed to sensory-sensory) arbitrary associations (Canavan et al., 1989a). Lesion studies in monkeys had provided only indirect evidence of a basal ganglia involvement by demonstrating an impairment following damage to the thalamic relays that convey information from the basal ganglia to the frontal cortex (Canavan et al., 1989b). Likewise, few electrophysiological data were available. Some, recorded in well-trained animals,

during learning (Boussaoud & Kermadi, 1997).

In this context, the present research was undertaken in order to specifically test the hypothesis of a pivotal role of the basal ganglia in the learning of new conditional visuomotor associations. In order to obtain converging evidence, the original plan was to combine the four approaches hitherto used in the field by combining: 1) electrophysiological recordings in the monkey striatum during learning of a visuomotor conditional task, 2) reversible inactivation of different striatal subregions in the same animals, 3) a neuropsychological evaluation of the ability of patients with Parkinson’s disease to learn such a task, and 4) a brain imaging study of the neuronal correlates of this type of learning in the normal human subject. The single-cell recordings were intended to demonstrate the existence of a learning-related plasticity within the striatum, and compare it to that described in PMd. The monkey and human lesion studies were aimed at providing further evidence that the basal ganglia are indeed necessary for normal learning. Finally, the imaging technique was seen as a unique tool to investigate different stages of learning, and evaluate how these two parameters affect activation in the basal ganglia and their anatomical connections, in particular in the frontal lobe. In all experiments, the subjects (monkeys, Parkinson’s patients and healthy human subjects) had to learn the same type of arbitrary associations, or rules, between visual cues and either hand or finger movements.

Because the reversible inactivation study has not yet been completed, the present report will focus on three experiments. In the first experiment, we recorded single unit activity in the striatum while monkeys either executed familiar associations (acquired prior to the recordings), or learned novel ones. The results identified strong learning-related changes of neuronal activity in the striatum, which were either transient (i.e. selectively occurring during early learning stages), or relatively long lasting (i.e. persisting through both early and late stages of learning). These results demonstrate for the first time that the learning-related changes that have been described earlier in PMd are also present in the striatum.

In the second experiment, advanced Parkinson's patients were tested on a series of tasks to determine the possible source of their difficulties in learning conditional associations. Their performance was assessed both with (ON) and without (OFF) dopaminergic treatment, and was compared to the performance of normal controls. We found that a subgroup of PD patients had marked difficulties to learn conditional associations in the OFF condition. This deficit was associated to poor use of a compensatory strategy (termed 'motor strategy').

associative learning using functional MRI. Learning-related activation was studied by contrasting the pattern of activation observed during the early phase of learning new associations (EARLY) to that observed either during the final stage of learning (LATE) or during execution of well-mastered associations (FAMILIAR). Both contrasts revealed brain activation in a network including the premotor regions, medially (anterior cingulate cortex and the presupplementary motor area) and laterally (PMd), as well as the dorsolateral (Brodmann's areas 9/46 - 10) and ventrolateral (Brodmann's areas 47/44) prefrontal cortex, the parietal cortex (intraparietal sulcus, precuneus), the right inferior temporal gyrus, and the cerebellum. Interestingly, subcortical regions, and more precisely, the striatum and mediodorsal thalamic nucleus, were found to be equally active during EARLY and FAMILIAR stages, but less importantly recruited during the LATE stage.

Taken together, the findings of these experiments not only confirm a role of the basal ganglia in the learning and use of arbitrary rules, but also improve our understanding of the dynamics of activity changes in the striatum and the precise source of the deficits related to dopamine depletion in Parkinson’s patients. In the following chapters, an Introduction on the anatomy and function of the basal ganglia will be presented, before describing data from each of the three experiments. The overall contribution of this research to current understanding of the basal ganglia involvement in nonstandard mapping will then be discussed in a final General Discussion section.

SECTION 1 -

The basal ganglia (BG) are the largest subcortical nuclei in the human brain. They form a functional system consisting of several structures. The naming of the BG has led to some confusion over the years as to which structures should be included within this description. Now, it is generally admitted, albeit not unanimously, that the BG include the caudate nucleus and the putamen (which are collectively referred to as the striatum), the globus pallidus, the substantia nigra and the subthalamic nucleus. These nuclei are heavily interconnected. Specifically, they are organized in functional loops, the cortico-basal ganglia-thalamo-cortical loops. Dysfunction of the BG, and of the functional loops they are involved in, leads to major motor disorders such as Parkinson's disease (characterized by hypokinesia) and Huntington's disease (characterized by hyperkinesia). There is, however, growing evidence that the BG are not important solely for the preparation, initiation and execution of complex automatic and voluntary movements, but contribute as well to non-motor, cognitive and motivational functions.

I. T

: H

The first clear identification of the ‘basal ganglia’ was published by the English anatomist Thomas Willis in 1664, in his basic foundational text on the anatomy of the central nervous system, Cerebri Anatomie, written in Latin (Figure A1, cf. Parent, 1986). The term ‘basal ganglia’ was not yet introduced. These subcortical structures were then denominated as the ‘corpus striatum’, and included the caudate nucleus, the putamen and the globus pallidus. Two characteristics drew attention to them. First, their central position in the brain suggested that they should play an important role. Second, massive ascending fibers projecting to them and descending fibers emerging from them raised the possibility that the BG might both receive all sensory modalities and initiate all motor acts.

By the 18th century, subsequent research shed light on the cerebral cortex, living the corpus striatum in the dark. Indeed, the attractiveness of the histological organization of the cortex, and the possibility of localizing higher mental functions drew many neurologists of both the 18th and 19th centuries to cortical research.

in 1872. During the 20th century, for the majority of the neuroanatomists, the term basal ganglia (also called basal nuclei) referred to the corpus striatum of Willis.

At the beginning of the 20th century, these structures began to gain importance once again with the discovery that their lesions often result in disorders of motor functions in humans (Wilson, 1914, see page 16). There were serious attempts to provide detailed comparative descriptions of the corpus striatum (Wilson, 1914; Cajal, 1911; Vogt & Vogt, 1920). Vogt & Vogt (1920) published descriptions of the connections between the thalamus and corpus striatum. This accumulation of data was accompanied by controversies regarding the list of structures composing the BG, apart from the corpus striatum of Willis. This lack of consensus explains the famous sentence of Thomas Thach: ‘the basal ganglia are no longer

mysterious now they are just confusing'. Parts of the thalamus, the amygdala, and the

claustrum, have all in turn been viewed as part of the BG, before the currently predominant view including the substantia nigra and the subthalamic nucleus emerged.

Over the 2nd half of the 20th century, the corpus striatum came progressively to be viewed as the major component of the "extrapyramidal motor system" (Parent, 1986), a system responsible for coordinating and integrating various aspects of motor behavior or body movements. Its “motor” role has been extensively studied. More recently, it has been implicated in various cognitive functions.

II. T

B

(

A2

)

From a functional and clinical point of view, the BG include the striatum (caudate and putamen) and the globus pallidus, together with two brain stem nuclei, the substantia nigra and the subthalamic nucleus (Carpenter, 1981), which are derivatives of the diencephalon and mesencephalon, respectively. Despite their phylogenitically and ontogenetically distinct origins, these brain stem nuclei are parts of the functional system that arise from the cortex, pass through the striatum, the pallidum and substantia nigra, the thalamus, and project back to the frontal cortex. The striatum constitutes the input stage of the BG. It receives information from virtually all cortical areas as well as from several subcortical areas, including most of the neuromodulatory systems. In this section, a description of the BG will be provided with special emphasis on the striatum.

1. The striatum

The term ‘striatum’ was first introduced by Vogt & Vogt in 1920 to refer to the telencephalic ensemble formed by the caudate nucleus and the putamen. It consists of the largest component of the BG and is considered to represent the first stage of neural computation in the BG.

The caudate nucleus is a large C-shaped structure located medial to the internal capsule. The term derived from a Latin word that means ‘having a tail’. It has an enlarged rostral component (head) that bulges into the lateral wall of the frontal horn of the lateral ventricle. The body in turn becomes further attenuated to form the tail which terminates at the amygdaloid nuclei. The body follows the lateral wall of the lateral ventricle. The tail occupies a position in the roof of the inferior (temporal) horn of the lateral ventricle. In essence, the caudate nucleus follows the curvature of the lateral ventricle.

The putamen is a shell-shaped structure situated medial to the cortex of the insula and surrounded laterally by the external capsule, medially by the lateral medullary lamina of the globus pallidus, and dorsally by the white matter of the corona radiata.

In primates, the putamen and caudate nucleus are incompletely separated by the internal capsule. The two nuclei form a homogenous component, sharing anatomical and cytological similarities (DeLong & Georgopoulos, 1981). They are continuous at the base of the hemisphere around the anterior limb of the internal capsule and are linked by scattered cells that bridge across the anterior limb of the internal capsule. The head of the caudate

nucleus and the putamen are connected by thin bridges of grey matter (pontes grisei caudatolenticularis). The name striatum or striate body is derived from the striated (striped) appearance of the internal capsule as it passes through these nuclei. In rodents, the two structures are not separated by the internal capsule, and are therefore often referred to as the caudate-putamen.

a. Anatomical subdivisions

The striatum is functionally divided into the dorsal and the ventral striatum. The dorsal striatum includes most of the caudate nucleus and the putamen while the ventral striatum comprises the medial and ventral parts of the caudate/putamen, the adjacent nucleus accumbens, and the striatal part of the olfactory tubercle. Allo- and periallocortical areas project principally to the ventral striatum, and neocortical areas project mainly to the dorsal striatum (e.g Lynd-Balta & Haber, 1994). It is on the basis of this regional organization that the dichotomy into limbic- vs. nonlimbic-related striatal regions has been introduced (Heimer & Wilson, 1975). Across this thesis, most of the work presented or cited will be related to the dorsal striatum.

b. Cytology

Unlike cortical cells, striatal cells are densely packed and do not exhibit any dominant configurations or laminations (Jones, 1984). However, as all other major central nervous system nuclei, the striatum is composed of both projection neurons and local interneurons corresponding to Golgi type I and type II cells, respectively, as first identified and denominated by DiFiglia et al. (1976). Still, contrary to most brain structures, the projection neurons greatly outnumber interneurons in the striatum. The ratio of projection neurons versus interneurons is approximately 9:1 in rodents, whereas it is 3:1 in primates (Graveland & Difiglia, 1985).

‰ Projection neurons: the medium spiny neurons

The striatum is primarily composed of projection neurons, originally described by Ramon y Cajal in 1911 (Graybiel et al, 1979, Kemp & Powell, 1971). They have a medium sized cell body (12-20 µm in diameter), which gives rise to 3-5 smooth primary dendritic branches, densely covered in spines (Kemp and Powell 1971; DiFiglia et al. 1976). Due to these morphological characteristics, the projection neurons have thus been termed "medium spiny" neurons. Furthermore, the axons of these neurons emit several collaterals, which arborize profusely and contact other spiny neurons (Kawaguchi et al., 1990). An example of these neurons is shown in Figure A3. Striatal projection neurons utilize GABA as their primary neurotransmitter (Smith et al. 1987). They also express a number of neuroactive peptides, such as substance P, enkephalin, dynorphin and neurotensin (Bolam et al., 1983). Not all of these peptides are found in every spiny neuron. The expression of these peptides seems to be related to the target nuclei of the spiny cell.

Neostriatal spiny neurons exhibit spontaneous fluctuations in membrane voltage which consist of transitions between two preferred potentials (Wilson & Groves, 1981), a relatively depolarized level referred to as the Up state (-55 mV) and a more polarized condition termed the Down state (-77 mV; see Wilson, 1993 for review). Action potentials are only generated from the Up state. The spiny cells are thus electrically quiescent in the absence of any extrinsic influence and require massive, relatively synchronous excitatory inputs to produce state transitions and spike triggering. As a consequence, they exhibit low firing rates (< 0.01 -0.5 Hz), and short duration extracellular action potential waveforms (Alexander & DeLong, 1985). The main striatal inputs, the cortical inputs, are glutamatergic, and the projection neurons have both non-NMDA and NMDA receptors (Kita, 1996). The striatum also receives dopaminergic inputs, and the projection neurons thus express dopaminergic receptors. There are two subtypes of DA receptors in the striatum. D1 receptors have an excitatory effect, and D2 receptors have an inhibitory effect (DiChiara, 1994).

‰ Interneurons

Two broad categories of interneurons have been identified based on their cell diameters: the giant aspiny interneurons and the medium aspiny interneurons (see Kawaguchi et al., 1995). The giant aspiny neurons contain choline acetyltransferase (Bolam et al. 1984; Phelps et al. 1985; Graybiel et al. 1986; DiFiglia, 1987). The medium aspiny neurons include the paravalbumin-containing GABAergic aspiny cells (Gerfen et al. 1985; Kita et al. 1990). Other subcategories have also been described such as the somatostatin/NOS (nitric oxide synthase) containing GABAergic apsiny cells (for a review, see Kawaguchi et al., 1995). For the purpose of this review, there will be only a brief description of the two principal subcategories of interneurons, namely the cholinergic-, and the parvalbumin-containing neurons.

The giant aspiny cholinergic interneurons are the best known interneurons. Ramon y Cajal first considered them to be projection neurons. These cells possess large spherical, oval or elongated cell bodies (approximately 20-35 µm in diameter in rat and primate) from which 2-5 smooth or sparsely spiny dendrites radiate (Bolam et al. 1984; Phelps et al. 1985; DiFiglia 1987). They are identifiable by their content in choline acetyltransferase, the most faithful marker of cholinergic neurons (DiFiglia, 1987; Phelps et al, 1985). These neurons are supposed to correspond to the physiologically defined tonically active neurons (TANs), so called because they fire tonically yet irregularly at 2-10 Hz (Kimura et al., 1984; Bolam et al., 1984, for a review, see Apicella, 2002). Their resting potential is relatively close to the spike threshold. Pharmacological blockade of spontaneous excitatory, inhibitory and neuromodulatory synaptic inputs to cholinergic interneurons did not influence spontaneous firing in vitro, demonstrating that these cells are tonically active in the absence of any input (Bennett and Wilson, 1999).

The paravalbumin-containing GABAergic aspiny cells (fast-spiking cells) exhibit spherical cell bodies (14-15 µm) and have axons with very dense collateral arborizations (Kita et al., 1990). They display immunoreactivity to GABA and/or its synthesizing enzyme glutamic acid decarboxylase (GAD) and also to paravalbumin, a calcium-fixating protein. This class of interneurons, embedded with gap-junctions, fire phasically at high frequency in response to cortical stimulation (Kita et al., 1990).

projection neurons. Like pyramidal cells in the cortex, they receive most of the inputs to the striatum and send almost all the efferent fibers. Hence, they provide synaptic input to other BG nuclei and, through local axon collaterals, contact interneurons and other spiny cells. The interneurons, despite their relatively small number compared to the projection neurons, have been shown to exert a powerful control on the activity of projection neurons in the striatum. Nonetheless, like projection neurons, interneurons receive direct inputs from cortical and other afferents to the striatum. Although there are numerous morphologically distinct classes of striatal cells, typically only two types of neurons are reported in unit recording studies. The first class refers to the projection neurons, which exhibit phasic increases in firing in response to cortical stimulation (Phasically Active Neurons, PANs). The second class corresponds to the cholinergic interneurons (Tonically Active Neurons, TANs).

c. Functional domains : matrix/striosome compartments

Neurochemical evidence has allowed to subdivide the striatum into two broad compartments, the striosomes (also called patches) and the matrix (Gaybiel & Ragsdale, 1978; Graybiel, 1995). These compartments were defined by the intensity of histochemical staining for acetylcholinesterase in cats and primates (Gaybiel & Ragsdale, 1978), and by heterogeneous distribution of µ opiate receptors in rodents (Herkenham & Pert, 1981). This compartmentalization is present both in the dorsal and ventral striatum, with the exception of the shell region of the nucleus accumbens (Voorn et al., 1989).

The striosomes, which occupy only about 15% of the striatum (Johnson et al., 1990), are rich in µ opiate receptors, neurotensin and AMPA receptors. They constitute a set of discrete modules with clearly defined boundaries. They are surrounded by a large matrix, which is rich in AChE, somatostatin and calbindin. Striosome-like domains have been identified in the matrix (Graybiel et al., 1994; 1995), and have been termed ‘matrisomes’. Individual cortical cells projecting to the matrix often form several small discrete arborizations of approximately the same size as those in the striosomes (Kincaid et al., 1998). This particular architecture provides the striatum with a discrete modular organization in a way that is analogous to the columnar structure of the cortex. Spiny neurons strictly obey the striatal compartment boundaries, with cells in the striosomes keeping their dendritic fields restricted to the striosomes and cells in the matrix having their dendritic fields contained within the matrix. The TANs are largely confined to the borders of the striosomes and the matrix. Given this preferential localization, these interneurons are believed to mediate

interactions between striatal projections of both compartments (Aosaki et al., 1995; Kawaguchi et al., 1995). These interneurons have been suggested to be recipients of direct cortical, thalamic as well as dopaminergic inputs (Wilson et al. 1990) and have been implicated in striatal plasticity.

In summary, the striatum is a heterogeneous structure, exhibiting different levels of anatomical and neurochemical organization.

2. The Globus pallidus (GP)

The globus pallidus is a wedge-shaped structure located between the putamen and the posterior limb of the internal capsule. It is situated medial to the putamen and is separated from it by a thin lamina of myelinated fibers, the lateral medullary lamina. A similar lamina, the medial medullary lamina, divides the GP into a lateral (or external) segment and a medial (or internal) segment. Thus, the GP is crossed by numerous myelinated fibers which explain its characteristic appearance in stained sections and from which derives its name ‘pale body’. The term ‘lenticular or lentiform’ nucleus is sometimes applied to the putamen and globus pallidus together because of their combined lens-shaped aspect in brain sections.

The globus pallidus is divided into three functional domains: the internal (GPi), the external (GPe) globus pallidus and the ventral pallidum (VP, the more anterior part of the GP, located under the anterior commissure). Although these domains are traversed by fibers, their neuronal populations are extremely similar, and for the most part morphologically indistinguishable (Carpenter, 1981). In humans, the GPe constitutes 70% of the total volume of the globus pallidus (Thorner et al., 1975). In non primates, the GPe and GPi usually have a larger separation and are referred to as the pallidum and entopeduncular nucleus, respectively.

There is a variety of neuronal types in the globus pallidus, but all are GABAergic neurons. The majority of them has a large ovoid body (20-60 µm in their long axis), with four to five long, thick and relatively smooth dendrites (Francois et al., 1984). The large dendrites can extend up to 1mm in length as illustrated in Figure A3. In rodents, it has been shown that the dendrites form a discoidal dendritic field and are disposed perpendicularly to striatal afferent axons (i.e. parallel to the lateral medullary lamina separating the globus pallidus from

the putamen). This positions the dendritic fields so as to intercept maximal numbers of striatal afferents (Park et al. 1982). GP cells are 100 times less numerous than spiny striatal neurons, which suggests a numerical convergence of striatal projections neurons on pallidal cells.

3. The substantia nigra (locus niger)

The substantia nigra is the largest single mesencephalic nucleus. It lies in the ventral tegmentum of the mesencephalon, forming an elongated nucleus that runs throughout the midbrain (Figure A2). It is divided into two components that have different connections and distinct neurotransmitters, a more ventral part with low cell density, the substantia nigra pars reticulata (SNr), and a dorsal part with high cell density, the substantia nigra pars compacta (SNc),. The latter is composed of large neurons exhibiting a characteristic black pigmentation; hence the origin of the structure's name ("black substance or locus niger"). Neurons of the SNc use dopamine as a neurotransmitter and project primarily to the striatum. Neurons in the SNr project principally to the thalamus (ventral anterior, ventral lateral and mediodorsal nuclei) but also to brainstem nuclei (superior colliculus, pedonculopontine nucleus) and use GABA as neurotransmitter. These neurons fire regularly and continuously at a very high rate (up to 100 Hz at rest; Chevalier & Deniau, 1990).

4. The Subthalamic nucleus (Luys Body)

The subthalamic nucleus is a biconvex structure located on the medial side of the internal capsule (Figure A2). It was discovered in 1865 by the French doctor Jules Bernard Luys, and was later named Luys body by August Forel, in recognition of its discoverer. Luys not only discovered the subthalamic nucleus, but he was also the first to think of this structure as being intimately linked to the BG. Among the BG neurons, the subthalamic neurons represent the only excitatory ones, using glutamate as their neurotransmitter.

In summary, the BG are the largest subcortical nuclei in the human brain. They form a functional system consisting of several structures: the striatum, composed of the caudate nucleus and the putamen, the globus pallidus (internal, external, and ventral segments), the substantia nigra (pars compacta and pars reticulata) and the subthalamic nucleus. From the morphological characteristics of the different components of the BG, two important features should be outlined. First, as illustrated for the PANs striatal neurons and GP neurons (Figure

A3), the BG neurons exhibit large dendritic fields and important axonal collaterals. Second, there is a dramatic decrease of cerebral tissue volume from the cerebral cortex to the striatum as well as within the BG structures. Yelnik and co-workers (2002) have made a computer-aided, three dimensional cartography of the BG (cf Figure A4). They found that the volume of the striatum is 12 times larger than that of the GPe, 20 times larger than that of the GPi and SNr, and 60 times larger than that of the STN. It has been proposed an estimated convergence of about 30:1 (rat) to 80:1 (monkey) for striatal projections onto their target neurons. Anatomically, the striatopallidal system is thus characterized by a considerable volumic, numeric, as well as geometric convergence. These features obviously denote an important and

III.

I

'

'

The BG are classically viewed as part of neural circuits that arise from the cortex, pass through the striatum, the pallidum and substantia nigra, the thalamus, and project back to the cortex, especially the frontal cortex (Figure A5). The striatum constitutes the input stage of the BG. The GPi and SNr constitute the principal output stages of the BG. The striatum receives projections from almost all cortical areas, as well as from subcortical areas, including most neuromodulatory systems. Among these afferent inputs, the projections arising from the cortex are by far the most prominent, and originate mainly from the ipsilateral cortex. Depending on the striatal target, they arise from neurons located in either supragranular and infragranular cortical layers (Gerfen, 1990). These cortical projections are of particular interest as they seem to impose upon the striatum a pattern of functional organization that is maintained throughout the BG, i.e. what is known as the BG loops or the cortico-basal ganglia-thalamo-frontocortical circuits. In addition to their close relationship with the frontal cortex, the BG nuclei send outputs to brainstem nuclei involved in motor control, including the superior colliculus, which controls axial orientation and saccadic eye movements.

1. The cortico-striatal projections : a funneling or a parallel processing?

The existence of a corticostriatal projection had been a somewhat contentious issue until convincingly shown by Glees (1944). Subsequently, several models have been proposed, suggesting either convergence or segregation of the information processing throughout the BG. I will review some of these models below.

a. Kemp & Powell's proposal

Early investigations with the Glees (Glees, 1944) and Nauta lesion-based techniques (e.g. Nauta & Mehler, 1966) have shown the presence of cortico-striatal fibers arising from the entire extent of the neocortex. Although Cajal (Cajal, 1911) considered corticostriatal fibers to be collaterals of corticofugal projections destined for lower centers, studies using horseradish peroxidase (HRP) have clearly demonstrated that these fibers, both ipsilateral and controlateral, arise from cell populations distinct from those that form the corticospinal, corticobulbar, corticopontine, corticorubral and corticothalamic systems (Jones et al., 1977a,b). In order to characterize the organization of the cortico-striatal projections, Powell and his co-workers made lesions in virtually every areas of the cortex of 47 rabbits (Carman et al 1963) as well as in monkeys (Kemp & Powell, 1970; 1971), and plotted the ensuing

degeneration on reconstruction of the striatum. The cortico-striatal fibers have been found to constitute a massive, topographically organized projection to the striatum with a relative degree of overlap. A mediolateral and anteroposterior topography was described, with the cortex of the frontal lobe being related to the anterior part of the head of the caudate nucleus and putamen, and the visual cortex at the occipital pole to the posterior part of the striatum. In the frontal lobe, the medial surface projects dorsally in the striatum, the lateral surface laterally and the orbital cortex medially. Thus, according to the view proposed by Kemp and Powell, the cortico-striatal projections followed the 'rule of proximity, each striatal region

receiving projections from the nearest overlying cortical area' (Parent & Hazrati, 1995a).

From these data, Kemp & Powell proposed that the BG serve to integrate diverse inputs from the entire cerebral cortex and to 'funnel' these influences to the BG output and to the primary motor cortex (Allen & Tsukahara, 1974; Evarts & Thach, 1969; Kemp & Powell, 1971; Nauta & Mehler, 1966). According to this view, there is “funnelling” from wide-spread cortical territories to narrower target areas in the thalamus. Thus, the BG could provide a route by which 'not only the sensory pathways but also the areas of the association cortex of

the frontal and parietotemporal lobes' could influence the motor cortex, allowing

convergence of the information relevant to the initiation and control of movement (Kemp & Powell, 1971). On the basis of these anatomical findings and the motor deficits observed after BG lesions, these structures were thought to project exclusively to motor cortical areas and to participate essentially to motor functions.

b. Alexander, Delong and Strick's proposal

The funneling model has been challenged by more recent data. First, it has been suggested that cortical projections to the striatum are topographically organized, in such a manner that non-adjacent, but functionally related regions, such as areas in the prefrontal and parietal cortices, project to close or even overlapping striatal sectors (Selemon & Goldman-Rakic, 1985; Flaherty & Graybiel, 1991). Second, the BG were found to send information not only to motor areas, but to various frontal regions as well.

In the early 1980's, DeLong and his coworkers suggested that the topographic mapping of cortical inputs provided functionally differentiated striatal subregions which in turn give rise to topographic, restricted projections to the GPi/SNr and thalamic nuclei, preserving the organization until the frontal cortex. This organization introduced the notion of parallelism (Delong & Georgopoulos, 1981; Delong et al, 1983; Kemp & Powell, 1970;

loop which links the sensorimotor and premotor cortex through the putamen, and an 'association or complex' loop passing through the caudate nucleus, which receives inputs from the association areas and return to the prefrontal cortex (Delong & Georgopoulos, 1981; Delong et al, 1983). In this model, the loops are relatively segregated and subserve distinct functional roles. The recognition that information originating from different parts of the cortex may remain segregated in parallel pathways that pass through the striatum to the pallidum or substantia nigra challenged the traditional view according to which the striatum serves as a kind of funnel trough which information from the cortex converges onto a limited number of output targets.

In 1986, the same group extended this new idea of segregated loops (Alexander et al., 1986, see Figure A6) by suggesting the existence of at least five loops, defined by their cortical origin: a motor loop originating in the supplementary motor area, an oculomotor loop originating in the frontal eye field, a dorsolateral prefrontal cortex (DLPF, area 46) originating loop, a lateral orbitofrontal cortex (LOF, area 12) originating loop, and a loop originating in the anterior cingulate and medial orbitofrontal cortices (AC/mOFC, areas 24 and 13) (Alexander et al., 1986). An additional feature of this scheme is that the loops are not only parallel but essentially closed, originating and terminating in the same frontal cortical region. The motor loop has received much attention because of its supposed involvement in movement-related disorders such as those observed in Parkinson’s disease.

Considerable anatomical and neurophysiological evidence supports the concept of a parallel BG organization. Hoover and Strick (1993) provided the most convincing evidence in experiments using attenuated herpes virus as a transneuronally transmitted tracer of connectivity. This view posits that the BG are in a position that enables them to influence frontal regions involved not only in motor functions, but also in higher executive functions such as planning, working memory, learning, and attention.

c. Parent's poposal (Figure A7)

In a similar simplified manner, Parent suggested the subdivision of the striatum into three functional domains: a sensorimotor, an associative and a limbic domain (Parent, 1990; Joel & Weiner; 1994), based on the topographical organization of the corticostriatal projections. In primates, the motor striatum comprises the dorsolateral postcommissural putamen and the dorsolateral region of the caudate nucleus. It is innervated by the primary motor cortex, premotor cortex, supplementary motor and lateral premotor area (Alexander & Crutcher, 1990; Alexander et al., 1990; Parent, 1990; Yeterian & Pandya, 1991; Selemon & Goldman-Rakic, 1985). This domain resembles the motor loop as defined by Alexander et al. (1986). The associative striatum comprises large parts of the putamen, rostral to the anterior commissure, and most of the head, body and tail of the caudate nucleus. It receives inputs from associative areas of the cortex, including areas 8, 9, 10 and 46 of the prefrontal cortex in the primate (Parent, 1990; Yeteran & Pandya, 1991). This domain resembles the striatal target of the oculomotor, the DLPF as well as the LOF loops, as proposed by Alexander and colleagues (1986). The limbic striatum comprises the nucleus accumbens and the most ventral parts of the caudate and putamen. It receives extensive inputs from limbic structures, such as the hippocampus and amygdala, as well as from prefrontal areas subserving limbic and autonomic functions, i.e the orbitofrontal cortex and anterior cingulate areas. This last domain resembles the striatal target of the AC/OFC loop as defined by Alexander and colleagues (1986).

2. The Nigrostriatal projections

The mesencephalic dopaminergic (DA) system is the largest dopaminergic system in the brain. The organization of DA neurons in rats and primates is generally similar. First described in rats using a fluorescence histochemical method (Dahlström & Fuxe, 1964) and subsequently in non-human (Felten et al., 1974) and human primates (Nobin & Björklund, 1973), this mesencephalic DA system is formed by three cell groups: the retrorubral area (RRA, group A8), the substantia nigra (almost exclusively the SNc and to some extent the SNr, group A9) and the ventral tegmental area (VTA, group A10).

The anatomical division of the DA cells is considered to reflect differences in their efferent projections as well as morphological and chemical characteristics. The loosely spaced neurons in the dorsal tiers, i.e the dorsal part of the SNc, the VTA and the RRA, display a strong immunoreactivity for clabindin d-28K and relatively low level of tyrosine hydroxylase

(Gerfen et al., 1985; Haber et al., 1995). The ventral tiers (ventral part of the SNc) includes two parts, a densocellular part, which lies dorsal to the SNr, and columns of dopaminergic neurons that penetrate deeply into the SNr (Joel & Weiner, 2000; Smith & Kieval, 2000). Unlike the dorsal tiers, the ventral tiers does not display immunoreactivity for clabindin d-28K (Gerfen et al., 1985; Haber et al., 1995; Agid et al., 1987; Graybiel et al., 1990; Prensa et al., 1999; 2000; Joel & Weiner, 2000). In rats, the DA neurons from the ventral tiers innervate preferentially the striosomes whereas DA neurons from the dorsal tiers innervate preferentially the matrix (Gerfen et al., 1987). This preferential distribution of the DA projections to specific compartments in the striatum is less clear in the monkey (Graybiel et al., 1987).

Recent reviews have summarized the DA inputs to the striatum according to its functional subdivisions, i.e. the associative, the motor and the limbic subdivisions (Haber & Fudge, 1997; Smith & Kieval, 2000). It seems that the sensorimotor striatum receives its main DA inputs from the cell columns in the ventral part of the SNc in primates. The limbic striatum receives different DA inputs arising from the VTA as well as from the dorsal part of the SNc. Finally, the associative sector of the striatum is innervated by a wide range of DA neurons located in the densocellular part of the ventral SNc. Five types of DA receptors have been described, the D1, D2, D3, D4 and D5. It seems that DA stimulation leads to an activation of the D1 and D5 receptors (previously grouped as D1-like receptors) and an inhibition of D2/D3/D4 receptors (previously grouped as D2-like receptors). Throughout the three sectors of the striatum, spiny neurons contain D1 and D2 receptors. A certain degree of co-localization of these two subtypes of receptors has been reported in the spiny neurons (Aizman et al., 2000). D3 receptors are also found in the limbic striatum. This receptor seems also to co-localize with D1 and D2 receptors. Some interneurons also expressed dopaminergic receptors. For instance, cholinergic interneurons have been found to express D2 and D5 receptors (Lemoine & Bloch, 1990).

Some studies have suggested that SNc and VTA dopamine neurons also innervate, although less massively, the globus pallidus particularly the internal segments, the ventral pallidum and the STN (Lindvall & Kjorkund, 1979; Cossette et al., 1999). Another non negligible source of DA inputs to the BG is the dentritic release of DA in the SNr, where dopaminergic receptors have also been identified (Mrzljak et al., 1996).

Thus, the interactions between the mesencephalic DA nuclei and the BG seem to be more important and more diffuse than previously believed.

3. The Thalamostriatal projections

In addition to the cerebral cortex, the thalamus constitutes another important source of excitatory inputs to the striatum (Parent, 1986; Wilson et al. 1983). The thalamostriatal projections were first demonstrated in humans by Vogt and Vogt in 1941 (Vogt & Vogt, 1941, Parent, 1986). These projections seem to be almost exclusively ipsilateral and they innervate the whole striatum including the nucleus accumbens (Parent, 1986).

The thalamus is composed of several nuclei, including : 1) the intralaminar nuclei, i.e. the centromedian and parafascicular nuclei (CM-PF), 2) the relay nuclei, namely the lateral nuclei subdivided into the ventrolateral (VL), anterolateral (VA) and lateroposterior (LP) nuclei, the mediodorsal (MD) nucleus and the pulvinar, and 3) the midline nuclei.

The most prominent projections to the striatum arise from the intralaminar nuclei (Powell & Cowan, 1954; 1956). Other thalamostriatal projections originate in the midline thalamic nuclei (paraventricular, paratenial, rhomboid and reuniens nuclei), the MD and to a lesser degree in the lateral and posterior thalamic groups (Nauta & Mehler, 1966; Mengual et al., 2000). These thalamostriatal projections are topographically organized. The midline thalamic nuclei, the MD and the PF project preferentially to the associative and limbic territories of the striatum, whereas the rostral intralaminar nuclei, the CM, the ventral nuclei and LP groups project preferentially to the sensorimotor territory of the striatum (Giménez et al., 1995; Nakano et al., 1990).

4. Amygdalostriatal projections

The amygdala is a heterogeneous structure including several nuclei, the basolateral nuclear group (BL), the corticomedial region, and the central nucleus, which are thought to play specific roles in emotional processing (see Rolls, 2000). Because the amygdala has been considered as a component of the limbic system, it has been suggested that its projections to the striatum were mainly directed toward the limbic part of the striatum.

These amydgalo-striatal projections are topographically organized (see for example, Kitai & Kitai, 1990). Electrophysiological studies suggested that these projections are excitatory (Noda et al., 1968). A recent study in non human primates reported that the amygdala projections are preferentially directed toward the shell of the nucleus accumbens (Fudge et al., 2002). The BL seems to be the source of all amygdaloid inputs to the limbic striatum outside the shell. These projections terminate mainly in the striosomes compartments

part of the central amygdaloid complex as well as from the BL. Furthermore, few fibers have been found to arise from the BL and to project to the associative striatum. However, projections from the amygdala to other sectors than the limbic one is still a matter of controversy (Krettek & Price, 1978; Russchen & Price, 1985). A potential confound across studies is the variable definition of the ventral striatum due to its lack of cytoarchitectural boundaries.

5. Other sources of striatal inputs

The neuromediator serotonin (5-HT) is present in relatively high concentrations in the striatum, where it is believed to act as an inhibitory transmitter (Miller et al., 1975; Olpe & Koella, 1977). It arises from the dorsal nucleus of the raphe, also known as the supratrochlear nucleus (Szabo, 1970). Serotoninergic fibers are thought to innervate the striatum as well as the substantia nigra and the globus pallidus (for a review, see Halliday et al., 1995). Sparse noradrenalin fibers originating from the locus coeruleus have also been identified in the striatum (Marien et al., 1994).

6. Integration by striatal neurons of different inputs

Excitatory, glutamatergic inputs from the cerebral cortex synapse almost exclusively with the spine heads and distal dendritic areas, whereas inputs from the substantia nigra pars compacta, the thalamus, or other intrinsic striatal neurons contact the proximal dendrites and somata (Kemp & Powell 1970, 1971). The latter inputs are therefore in a crucial position to modulate or inhibit cortical influences. Thus, spiny projection cells are recipients of synaptic inputs from an extremely diverse collection of axons arising from both extrinsic and intrinsic sources as illustrated in Figure A8.

Striatal interneurons, particularly the cholinergic and the somatostatin-containing ones, also receive a very diverse synaptic input. However, one clear anatomical difference between the interneurons, at least the cholinergic and GABAergic cells, and the spiny projection neurons is the spatial distribution of their inputs. As stated above, excitatory inputs are directed to the distal regions of spiny projection cells whereas interneurons receive excitatory inputs on the proximal dendrites and somata. This anatomical arrangement coupled with the profound differences in the electrical properties of interneurons indicates that the regulation of action potential generation in interneurons is likely to differ dramatically from that in spiny cells.

IV. O

'

'

The GPi and the SNr represent the main output nuclei of the BG. They send their projections to the thalamus, the superior colliculus and to the premotor nuclei of the brainstem.

It has long been suggested that the GPi is innervated by the motor striatum while the SNr is innervated by the associative striatum. But another view, elaborated by Alexander et al (Alexander et al., 1990; Alexander & Crutcher, 1990; Kawaguchi et al., 1990), suggested that each striatal region innervates both GPi and SNr. It seems that the functional segregation of the corticostriatal projections is largely maintained trough the circuitry of the BG (Alexander et al., 1986; Parent & Hazrati, 1995a,b) and through the pallidothalamic projection (Sidibé et al., 1997). The ventrolateral two thirds of the GPi, which receive inputs from the sensorimotor striatum, project to the VL and the central part of the CM. The regions of the GPi innervated by the associative and limbic striatum project to the parvocellular VA and the rostral part of PF (Sidibé et al., 1997). The VP, which is mostly innervated by the limbic striatum, projects modestly to the most medial magnocellular part of the mediodorsal nucleus (Haber et al., 1993).

The rostral nuclei of the ventral thalamus (VA, VL, VM) territories innervated by the BG outputs widely overlap with the thalamic territories projecting to the striatum; whereas more restricted areas of overlap are visible in the rostral and caudal intralaminar nuclei (CM-PF), which is the source of the major thalamic input to the striatum (Parent & Hazrati, 1992; 1993). The various thalamic nuclei send in turn projections to the frontal cortex, hence 'closing the loop'. In addition, other BG components send projection to the thalamus. For example, the GPe projects to the reticular thalamic nucleus (Hazrati & Parent, 1991).

Interestingly, it was recently demonstrated that the frontal lobe is not the unique and 'privileged' indirect BG target. Strick and colleagues, using retrograde transneuronal transport of herpes virus type 1, elegantly demonstrated projections to specific areas of the inferotemporal (Middleton & Strick, 1996) and posterior parietal cortices (West, Lynch and Strick, unpublished observations). Moreover, GPi as well as specific territories in the striatum send direct inputs to the pedonducopontine tegmental nucleus (PPN) (Nauta & Melher, 1966; Parent & Hazrati, 1995b). These projections to the PPN have long been ignored in the current model of the BG organization (Parent & Cicchetti, 1998).

V. B

:

On the basis of anatomical findings, Albin and co-workers (1989) proposed two segregated feedfoward pathways from the striatum to the GPi/SNr1 (Figure A9). The direct

pathway is formed by a projection from the striatum to the GPi/SNr, and then to the thalamus.

The indirect pathway projects to the GPi/SNr complex via the GPe and the STN. Early retrograde labeling investigations have supported the idea of distinct striatofugal projections arising from separate neuronal populations in the striatum (Parent et al., 1984; Selemon & Goldman-Rakic, 1990; Flaherty and Graybiel, 1993). These studies demonstrated that the majority of spiny neurons which project monosynaptically to the output nuclei of the BG, specifically the GPi and SNr, contain substance P and dynorphin and express the D1 dopamine receptor (Gerfen et al. 1990). The neurons projecting to the output structures through the Gpe and the STN express enkephalin and D2 dopamine receptors (Gerfen et al. 1990).

Dopamine modulates the activity of striatal neurons that give rise to the direct and indirect pathways, a modulation which depends on the type of receptor involved. D1 receptors have an excitatory effect, whereas D2 receptors have an inhibitory effect (Chiara, 1994). Thus, dopamine allows the direct and indirect pathways to counterbalance each other. Furthermore, the two pathways have antagonistic effects on the output structures: the direct pathway sends an inhibitory output to the GPi/SNr, whereas the indirect pathway results in excitatory effects, eventually promoting activation of the frontal cortex and action.

The main strength of this model (i.e. imbalance between the activity in the direct and indirect pathways) lies in its capacity to account for pathophysiological mechanisms of both hypokinetic and hyperkinetic movement disorders (DeLong, 1990, Albin et al., 1989). Converging evidence suggests that in Parkinson's disease, the loss of neurons in the nigrostriatal dopamine-containing pathway leads to an activation of striatal outputs to the GPe, and to an inhibition of striatal projections to the GPi and the SNr (see Albin et al., 1989). The model predicts that inhibition of the GPe neurons release the STN from its tonic inhibition by the GPe. Increased activity in the STN, the only region of the BG to contain

1 _{In 1966, Nauta & Mehler already provided evidence of separate projections of the internal and}

excitatory projection neurons, contributes to the increased output from the GPi and SNr to which it projects, resulting in the inhibition of the thalamus projections to the cortex (Alexander & Crutcher, 1990). This hypothesis is supported by the findings that STN and GPi firing rates are increased in PD (DeLong, 1990). Moreover, it has been shown that inactivation of these nuclei can alleviate the motor symptoms in Parkinsonian animals (Bergman et al., 1990) and human patients (Benabid et al., 2000).

VI. I

:

-

The BG models have gained considerable clinical relevance because of their importance in guiding drug development and new surgical approaches. With time, however, the shortcomings of these models have become apparent, necessitating revision and updating.

1. The striatum and the GPi/SNr complex: input and output structures, respectively?

Like the striatum, the STN also receives direct excitatory inputs from motor, premotor and prefrontal areas of the cortex (Hartmann-von Monakow et al., 1978; Nambu et al., 1996, Maurice et al., 1998). In primates, the projections arising from the primary motor cortex are the most important. Some cortical neurons innervate both the STN and the striatum, particularly in the prefrontal cortex (see Parent & Hazrati, 1995b). These direct cortical inputs to the STN innervate the entire STN with a mediolateral topography (Afsharpour 1985; Camteras et al 1990, Parent & Hazrati, 1995b). Therefore, the STN can also be considered as an input structure through which cortical information is transferred to output nuclei of the BG (Kitai & Deniau, 1981; see for review, Joel & Weiner, 1997, Smith et al., 1998).

Although the GPi/SNr complex represents the major output structure of the BG, direct projection from the GPe to the thalamus as well as to the PPN has been also described (Nauta & Melher, 1966; Parent & Hazrati, 1995a,b).

2. Direct/indirect model?

Although many data corroborate the Albin-Delong model, some recent findings seem incompatible with this current dual model of the BG. Indeed, neurons in the BG show extensive collateral connectivity (Parent et al., 2000). The results of the single-cell labeling studies (see also Parent et al., 1995) have revealed an abundance of striatal projection neurons with highly collateralized axons that provide branches to two or three of the striatal recipient structures. Such a high degree of axonal collateralization allows striatal neurons to send efferent copies of the same information to virtually all striatal targets and additional internal and external projections (Bolam et al., 2000).

Furthermore, D1 and D2 receptors co-localize on striatal neurons (Aizman et al., 2000), which suggest that all striatal neurons that project to the GPi could also project to the GPe (Wu et al., 2000). Wu and his co-workers (2000), and Parent & Parent (2002) showed that virtually all striatofugal axons send collaterals to the GPe and none project exclusively to GPi/SNr. They divided striatal spiny neurons into three subtypes, based on anatomical data. Type I medium spiny neurons that project solely to the GPe. Type II medium spiny neurons project to both the GPe and SNr. Type III medium spiny neurons project to the GPe, GPi and SNr (Wu et al., 2000). Thus as concluded by Parent & Hazrati (1995b) the 'GPe cannot be

considered as a simple relay structure in the indirect pathway, instead, it appears to be a major integrative structure that can affect virtually all components of the basal ganglia'.

Moreover, lesions to the GPi not only ameliorate the hypokinetic clinical characteristics of PD, but also alleviate hyperkinetic disturbances, and lesions of the thalamus do not lead to PD-like motor symptoms (Obeso et al., 2000), as opposed to what can be predicted by the model. Thus, currently accumulating evidence is challenging the classical Albin-DeLong model.

3. Information processing in the basal ganglia: feedfoward/feedback -parallelism/convergence?

Despite the above considerations, the striatum could still be viewed as an important recipient structure. The current debates concern principally the information processing through the BG. Indeed, until recently the Albin-DeLong model has been widely accepted. According to this view, information processing in the BG followed essentially a feedfoward route. However, considerable internal BG feedback loops have been identified. For example, the GPe is reciprocally connected with the striatum and the STN. The striatum is also reciprocally connected with the SNc and some thalamic nuclei, particularly the CM/PF complex. Furthermore, the dopaminergic inputs arising from the SNc do not only influence the striatum but several other BG components, such as the GPi and the STN. Another important connection has long been ignored in the current model of the BG organization, the connection with the PPN (for a review, see Winn et al., 1997). This brainstem structure, containing cholinergic and non cholinergic neurons, receives direct input from striatum. This structure is reciprocally connected with the SN, GP and the STN, and is directly involved in the basal-ganglia-mediated control of behavior (Nauta & Melher, 1966; Winn et al., 1997). For instance, it has been suggested that the striatum could desinhibit the control of descending PPN influences on medullary and spinal targets.